Animate17 October 2026

How to turn an AI image into a video clip

Your image is the first frame; your prompt is the motion. Comparing MiniMax H3 Max, H3 Max Turbo and Gemini Omni Flash.

By Anand Murugan, who makes Chitra Studio

I had an AI generated image of a Bharatanatyam dancer on a temple stage, and I wanted five seconds of her dancing. Image-to-video models do exactly this: they take your picture as the first frame and move it the way you describe.

Whether the clip is good comes down to two things: what you write, and which model you send it to. Here is what I learned from the clips in my own projects.

Animating a still

Animate in Chitra Studio. While the model works, the picture glows and a timer shows which model is animating it. The clip at the end plays with the sound the model made.MiniMax H3 Max1080p · 21 s · 112 creditsMy Prompt"The camera holds still. The Bharatanatyam dancer turns smoothly on the stage, her arms and posture transitioning fluidly before settling firmly into a new classical pose. Throughout the movement, the flames of the brass oil lamps in the background continuously flicker, casting subtly shifting warm light around her."

Open the picture and choose Animate.
Describe what moves and what the camera does. One line is enough.
If you like, press Improve with AI to have that line written out in full (1 credit). Undo puts your own line back.
Choose the model, the length and the resolution. The price is shown beside the Animate button before you press it.

Write the motion, not the picture

The picture already shows who is there, where they are and what the light is like. The prompt only needs to say what changes. Good prompts do some or all of these:

Say what the camera does. "The camera holds still", "pushes in very slowly", "tracks backward with them". Leave it out and the model decides for you.
Ask for one main action. Five seconds is one gesture: drawing a sword, lighting a candle, a turn.
Name the small things that should move. Steam, lanterns, falling snow, candle flames. These are what make a clip feel alive.
Say what should not move, when it matters, as in the café clip below.

Improve Prompt with AI does much of this for you. For the samurai below, it put the camera first, kept one action and added a detail, his breath in the cold, that the model then drew.

Three models, real clips

These are clips from my own projects, so each model animated a different picture: a feel for each, not a strict test. Times are how long each took; prices are what a five-second clip costs in Chitra Studio today.

Gemini Omni Flash1080p · 94 s · 105 creditsMy Initial Prompt"he slowly draws his katana, snow keeps falling, the sound of wind and steel, camera holds still"Improve Prompt with AI wrote"The camera holds completely still. Snow continues to fall steadily through the grove. In a slow, deliberate motion, the samurai draws his katana from the scabbard, sliding the bare blade smoothly into the open air while his faint breath plumes in the cold."

MiniMax H3 Max768p · 16 s · 56 creditsMy Initial Prompt"she sings with one arm raise, the gondola glides forward, the water ripples"Improve Prompt with AI wrote"The opera soprano raises one arm dramatically as she continues singing, her chest expanding with the aria. The gondola glides smoothly forward across the Grand Canal, sending gentle ripples parting across the sparkling water. The camera tracks steadily backward at a low angle, matching the boat's slow, graceful pace."

MiniMax H3 Max Turbo768p · 15 s · 28 creditsMy Initial Prompt"steam rises from the espresso cup, morning light shifts slightly, nothing else moves"Improve Prompt with AI wrote"The camera remains completely stationary. Wisps of steam gently rise from the espresso cup, swirling slowly upward into the air. The morning light shifts subtly across the frame at a slow, calm pace. Absolutely nothing else moves."

MiniMax H3 Max Turbo768p · 16 s · 28 creditsMy Initial Prompt"the vendor slowly scoops spice into a paper cone, the lanterns sway gently, light flickers on the spice mounds, camera pushes in very slowly"Improve Prompt with AI wrote"The camera pushes in very slowly. The vendor slowly scoops spice from a mound into a paper cone. Overhead, the brass lanterns sway gently, causing the warm light and geometric shadows to flicker subtly across the spice mounds as the evening haze drifts through the air."

MiniMax H3 Max Turbo is the cheapest at 28 credits for five seconds at 768p, and the quickest, at about 15 seconds. I use it for quiet motion: steam, lanterns, light, a slow push in.
MiniMax H3 Max costs twice as much (56 credits at 768p, 112 at 1080p). I use it for people: the singer's gesture, the dancer's turn.
Gemini Omni Flash was the slowest here, 50 to 94 seconds, and costs about the same as H3 Max at 1080p. I used it for bigger single actions: drawing the sword, a dancer landing a leap.
All three made their own sound to go with the picture. Press the speaker button on a clip to hear it.

Same picture, different words

Small changes in the prompt give you a different clip. These two are the same still of a dancer mid-leap, both on Gemini Omni Flash. My initial prompts differ by two words, "and lands", and Improve Prompt with AI took each in a different direction.

My Initial Prompt"the dancer completes her leap and performs a dance move"Improve Prompt with AI wrote"The dancer lands softly from her leap onto the wooden stage floor, her silk dress swirling and settling with the impact. Absorbing the descent, she flows seamlessly into a low, sweeping turn, extending her body outward in a controlled contemporary gesture. The camera holds at eye level, subtly reframing to keep her centered within the spotlight."

My Initial Prompt"the dancer completes her leap and lands and performs a dance move"Improve Prompt with AI wrote"The dancer descends gracefully from her leap, landing softly on the polished stage floor. Absorbing the momentum, she sweeps into a fluid, expressive pirouette, her dress swirling around her before extending an arm outward into a low, poised pose. The camera holds at eye level, panning smoothly to maintain focus on her movement."

Each was four seconds at 720p, 56 credits. If the motion matters, run a prompt twice, or word it two ways, and keep the one you like.

Tips

Draft cheaply, then finish. Try your prompt on H3 Max Turbo at 768p first. When the motion is right, run it on a bigger model or at a higher resolution.
Leave room in the frame. If someone is going to raise an arm or draw a sword, it helps if there is space in the picture for it.
One beat per clip. Five seconds is short, so ask for one main action and let the small things fill the rest.
Listen before you replace the sound. These models make their own audio. If you want something else, Add Sound in Chitra makes new sound for a clip.

Animate your own pictures

Chitra Studio is a Mac app for making and organising AI images, video and sound with more than 60 models. Every price is shown before you generate.

Download for Mac14-day free trial · 50 free credits · no card